Chapter 4.2 · The application and the answer

Mortgage denials

Lenders denied 13.8% of the home-purchase applications they decided in 2024 in the most deprived tenth of tracts and 7.8% in the least deprived, and in the tracts that have published rates for both groups, 7.4% of all tracts, non-Hispanic Black applicants were denied 10.0 percentage points more often than non-Hispanic white applicants.

Author: Vishal Singh
Affiliation: NYU Stern
Published: 9 October 2026

Lenders denied 278,451 of the 3,097,390 home-purchase applications that they decided in 2024, a rate of 9.0% (95% CI 8.7% to 9.3%), pooled over 76,072 census tracts in every state and DC. The denial rate was 13.8% (95% CI 13.3% to 14.3%) in the most deprived tenth of tracts and 7.8% (95% CI 7.2% to 8.4%) in the least deprived. In the 5,657 tracts that have published rates for both groups, which are 7.4% of all tracts, it was 16.2% (95% CI 15.7% to 16.8%) for non-Hispanic Black applicants and 6.2% (95% CI 6.0% to 6.5%) for non-Hispanic white applicants.

The data are tract totals from the Home Mortgage Disclosure Act, which requires lenders above size thresholds to report every application and what became of it. A denial rate is a share of applications that reached a decision. It leaves out people who never applied, applications that were withdrawn, and everything a lender weighs about the applicant: credit history, debts, savings and the size of the loan against the value of the home. The chapter describes where and for whom applications ended in a denial. It does not measure whether a lender weighed an applicant's race in a decision, and the gaps below are not estimates of discrimination.

The chapter cuts the data three ways: tracts ranked by deprivation, applicant groups within tracts, and the grades that the Home Owners' Loan Corporation gave neighborhoods in the 1930s. Chapter 4.1, Home value, income and redlining compares those grades with life expectancy and home values, and Chapter 4.3, Work, wages and access to jobs covers the earnings that stand behind an application.

What a denial rate counts

The file holds first-lien loans on site-built homes with up to four units that applicants intend as a principal residence, reported for 2024 by lenders above the HMDA reporting thresholds. It splits each tract by loan purpose. Home-purchase applications were denied 9.0% of the time, refinance applications 20.7% and cash-out refinance applications 35.8% (Figure 1). The chapter uses home-purchase applications throughout. The rates differ by purpose, so a mix of purposes would let the mix of loans move a place's rate, and only the home-purchase file has enough applications by group: 8,232 tracts have a published rate for non-Hispanic Black applicants, against 1,155 for refinance loans and 2,938 for cash-out loans. Puerto Rico's 637 tracts with a published cell are set aside. 69% of those with a deprivation rank sit where the most deprived tenth of the other tracts begins, and the pooled rate there is 19.1% (95% CI 18.3% to 20.0%) on 11,409 decided applications. Including them would move the most deprived tenth from 13.8% to 14.1% and the overall rate by 0.04 points.

Figure 1

Decided home-purchase applications were denied 9.0% of the time, against 20.7% for refinance and 35.8% for cash-out refinance loans

Denied share of decided applications, tracts of the 50 states and DC with a published cell, 2024, by loan purpose

Loan purpose	Tracts with a published cell	Decided applications	Denial rate (%)	Lower limit of the 95% interval (%)	Upper limit of the 95% interval (%)	Tracts with a published rate for Black applicants
Home purchase	76,072	3,097,390	9.0	8.7	9.3	8,232
Refinance	36,779	440,105	20.7	20.2	21.2	1,155
Cash-out refinance	46,525	616,812	35.8	35.3	36.4	2,938

Show data

Data for Decided home-purchase applications were denied 9.0% of the time, against 20.7% for refinance and 35.8% for cash-out refinance loans
conf_high	conf_low	decided	purpose	rate	tracts	tracts_black
9.30	8.70	3097390	Home purchase	9	76072	8232
21.20	20.20	440105	Refinance	20.70	36779	1155
36.40	35.30	616812	Cash-out refinance	35.80	46525	2938

The home-purchase file has the lowest denial rate and by far the most tracts with a published rate for Black applicants, 8,232 against 1,155 for refinance and 2,938 for cash-out loans. The chapter uses home-purchase applications from here on. The rate is denied applications divided by applications originated or denied, pooled over tracts, with a 95% interval from a bootstrap over counties.Note: Each row counts tracts whose all-applicant cell for the purpose has 10 or more applications, a withholding rule of this book's staging step that counts applications of every action taken. Cells under that floor are missing, never zero, and leave both the denied and the decided counts. Tracts differ across rows because the floor binds more often for refinance and cash-out loans.Note: The file covers first-lien loans on site-built homes with up to four units that applicants intend as a principal residence, from lenders above the HMDA reporting thresholds. Loan type (conventional, FHA, VA or USDA) is not in the tract file. Puerto Rico is left out.Source: HMDA Loan/Application Register, 2024, tract totals of decided applications; Area Deprivation Index, 2023; full counts, pooled rates

The rate is denied applications divided by decided applications, those that the lender originated or denied. Applications that the applicant withdrew or left incomplete, loans approved but not accepted, and loans bought from another lender are outside it. A borrower who was discouraged before applying, or steered to a different loan, never enters it. Every rate here is pooled, with denied and decided counts summed over tracts, and the median tract decides only 31 applications, so a single tract's rate is noisy. Intervals are 95% intervals from a bootstrap that resamples whole counties.

When the file was staged, any cell with fewer than 10 applications, whatever their outcome and purchased loans included, was suppressed, a rule of this book and not an HMDA flag. A suppressed cell is missing, never zero, and leaves both the denied and the decided counts. On that rule 5,999 tracts have no all-applicant home-purchase cell, each holding at most 9 applications. A cell for one group is withheld far more often, because the group itself needs 10 applications in the tract. Each group's rate therefore comes from its own set of tracts. The floor counts applications of every outcome, so a published cell can rest on few decided applications: 4,865 of the 76,072 tracts decided fewer than 10. Leaving out tracts below 30 decided applications lowers the overall rate from 9.0% to 8.5% (95% CI 8.2% to 8.8%).

HMDA records an applicant's ethnicity and race as the applicant reports them, with fallback rules when the applicant does not answer. The tract file keeps four groups: applicants who are white and not Hispanic or Latino, Black or African American and not Hispanic or Latino, Hispanic or Latino of any race, and Asian and not Hispanic or Latino. The chapter calls them non-Hispanic white, non-Hispanic Black, Hispanic and non-Hispanic Asian applicants. Applicants who are American Indian or Alaska Native, Native Hawaiian or Pacific Islander, of two or more races, joint, or of unavailable race count in the all-applicant rate only, so the four groups do not add up to it. Loan type (conventional, FHA, VA or USDA) is not in the tract file.

Denial rates rise with deprivation after the third tenth

Ranking the 75,978 tracts that have a deprivation rank by the Area Deprivation Index national rank, and cutting them into tenths with near-equal numbers of tracts (tracts tied on the rank stay in one tenth), gives a denial rate of 7.8% (95% CI 7.2% to 8.4%) in the least deprived tenth, 7.7% in the second and 7.6% in the third. The differences among those three tenths are within sampling error. After the third tenth the rate rises at every step, to 13.8% (95% CI 13.3% to 14.3%) in the most deprived tenth, which is 6.0 (95% CI 5.2 to 6.8) points above the least deprived tenth and 6.2 (95% CI 5.6 to 6.8) above the third (Figure 2). The 94 tracts with no rank are left out of this and the next two sections. The last step, from the ninth tenth to the tenth, is the largest.

The rank is built partly from income and housing costs, so the curve mixes where an applicant lives with how much money applicants there have. The file reports both. The median applicant income and the median loan amount of each tract, averaged over the tracts of a tenth and weighted by decided applications, are $64 thousand (95% CI $63 thousand to $65 thousand) and $162 thousand (95% CI $158 thousand to $167 thousand) in the most deprived tenth, against $279 thousand (95% CI $266 thousand to $292 thousand) and $827 thousand (95% CI $774 thousand to $879 thousand) in the least deprived. The least deprived tenth holds the largest loans. The tract file does not say whether loan size, income or credit history sits behind a denial, so the curve describes where applications are denied and not why.

Figure 2

The denial rate was 7.8% in the least deprived tenth of tracts and 13.8% in the most deprived, and it is level across the three least deprived tenths

Denied share of decided home-purchase applications, 2024, against the tract's Area Deprivation Index national rank: sampled tracts with at least 30 decided applications as dots and pooled rates for tenths of tracts as the line

Loading data

The line is flat across the first three tenths and rises at every step after that, ending 6.0 percentage points above the first tenth, while single tracts scatter widely around it. Dots are a random sample of 2,000 of the 39,697 qualifying tracts, each resting on a small number of applications (the median tract decides 31); the line plots the denied share of all decided applications in each tenth of tracts, with 95% intervals from a bootstrap over counties.Note: Dots are drawn from the tracts that decided at least 30 applications, with a fixed seed (stated in the methods note). The pooled rates use every tract with a published cell and a rank, 75,978 tracts, in tenths that hold near-equal numbers of tracts. A tract with no rank is left out (94 tracts).Note: The Area Deprivation Index is a national percentile, 1 least and 100 most deprived, and it understates deprivation in high-cost cities. Rates are not adjusted for the applicant's credit, debts, loan size or loan type, which the tract file does not hold.Source: HMDA Loan/Application Register, 2024, tract totals of decided applications; Area Deprivation Index, 2023; full counts, pooled rates

Single tracts scatter widely around the tenth means. One denial moves the rate of a tract that decides few applications by several points, and the dots in the figure come only from tracts that decided at least 30.

Black and Hispanic applicants are denied more often than white applicants across the deprivation range

Over the tracts where each group's cell is published, non-Hispanic white applicants were denied 6.5% (95% CI 6.3% to 6.6%) of the time, non-Hispanic Asian applicants 8.9% (95% CI 8.4% to 9.4%), Hispanic applicants 13.1% (95% CI 12.4% to 13.9%) and non-Hispanic Black applicants 17.0% (95% CI 16.4% to 17.7%). Each rate rests on a different set of tracts, so that comparison mixes places with groups. Within tenths of tract deprivation, the Black and Hispanic rates are above the white rate in every tenth drawn, and so is the Asian rate in the first eight (Figure 3).

The white curve is U-shaped. It is 6.7% (95% CI 6.0% to 7.5%) in the least deprived tenth, 5.4% in the third and 10.4% in the most deprived. The other three groups rise with deprivation: the Black rate from 14.6% in the third tenth to 20.3% in the most deprived, the Hispanic rate from 11.8% to 14.2%, and the Asian rate from 7.8% in the least deprived tenth to 12.9% in the eighth.

Figure 3

In every tenth of tracts with enough applications, Black and Hispanic applicants were denied more often than white applicants

Denied share of decided home-purchase applications, 2024, by applicant group and tenth of tracts on the Area Deprivation Index (1 = least deprived), points on at least 100 tracts

Denied share of decided applications (%)

Show data

Data for In every tenth of tracts with enough applications, Black and Hispanic applicants were denied more often than white applicants
conf_high	conf_low	estimate	series	tenth	tracts
7.47	6.01	6.72	Non-Hispanic white	1	4805
5.85	5.24	5.53	Non-Hispanic white	2	5356
5.65	5.17	5.41	Non-Hispanic white	3	5868
5.79	5.34	5.56	Non-Hispanic white	4	6305
5.86	5.42	5.64	Non-Hispanic white	5	6150
6.51	6.03	6.28	Non-Hispanic white	6	6079
6.90	6.49	6.69	Non-Hispanic white	7	6079
7.50	7.08	7.29	Non-Hispanic white	8	5995
8.55	8.08	8.31	Non-Hispanic white	9	5763
10.77	10.10	10.43	Non-Hispanic white	10	4227
16.80	13.85	15.46	Non-Hispanic Black	2	439
15.56	13.77	14.64	Non-Hispanic Black	3	633
16.86	14.74	15.80	Non-Hispanic Black	4	910
17.34	15.54	16.45	Non-Hispanic Black	5	1022
18.32	15.95	17.18	Non-Hispanic Black	6	1099
18.03	15.91	16.97	Non-Hispanic Black	7	1075
18.60	16.31	17.44	Non-Hispanic Black	8	973
19.54	17.20	18.38	Non-Hispanic Black	9	902
21.58	19.02	20.30	Non-Hispanic Black	10	1081
14.04	10.71	12.35	Hispanic (any race)	1	389
13.35	10.85	12.01	Hispanic (any race)	2	1376
12.59	11.07	11.85	Hispanic (any race)	3	1847
13.28	11.39	12.31	Hispanic (any race)	4	2307
14.32	12.51	13.46	Hispanic (any race)	5	2242
14.49	12.39	13.43	Hispanic (any race)	6	2143
14.89	12.78	13.83	Hispanic (any race)	7	1874
15.46	13.05	14.21	Hispanic (any race)	8	1536
15.24	12.83	14	Hispanic (any race)	9	1285
15.74	12.86	14.18	Hispanic (any race)	10	1045
8.74	7.25	7.84	Non-Hispanic Asian	1	2241
9.19	7.98	8.51	Non-Hispanic Asian	2	1661
9.69	8.02	8.85	Non-Hispanic Asian	3	1243
10.36	8.83	9.55	Non-Hispanic Asian	4	938
11.42	9.05	10.17	Non-Hispanic Asian	5	626
11.64	9.10	10.40	Non-Hispanic Asian	6	394
14.02	10.09	11.92	Non-Hispanic Asian	7	237
14.47	11.01	12.85	Non-Hispanic Asian	8	103

The lines for Black and Hispanic applicants lie above the line for white applicants in every tenth drawn, and the white line dips in the middle tenths and rises at both ends. Each line pools the denied and decided applications of one group over the tracts where its cell is published, with 95% intervals from a bootstrap over counties.Note: Groups are the applicant's reported ethnicity and race as HMDA derives them: non-Hispanic white, non-Hispanic Black, Hispanic of any race and non-Hispanic Asian. Other and unreported applicants count in the all-applicant rate only, so the four groups do not add to it.Note: A group's cell is published only in tracts with 10 or more of its applications, so each line covers a different set of tracts, and the sets for Black, Hispanic and Asian applicants are smaller than the set for white applicants. A point is drawn when it rests on at least 100 tracts; the Black line starts at the second tenth and the Asian line stops at the eighth.Source: HMDA Loan/Application Register, 2024, tract totals of decided applications; Area Deprivation Index, 2023; full counts, pooled rates

A point is drawn when it rests on at least 100 tracts. That drops the Black rate in the least deprived tenth and the Asian rate in the ninth and tenth, where few tracts have 10 or more applications from the group.

The gap within a tract shows no trend with deprivation or applicant income

Comparing groups across tracts mixes who lives where with how applications fare. A narrower comparison keeps only tracts where both groups have a published cell and sets each group's pooled rate in those tracts side by side. In the 5,657 tracts that hold both Black and white cells, Black applicants were denied 16.2% (95% CI 15.7% to 16.8%) of the time and white applicants 6.2% (95% CI 6.0% to 6.5%), a gap of 10.0 (95% CI 9.5 to 10.4) points, or 2.6 (95% CI 2.5 to 2.7) times the white rate (Figure 4). Hispanic applicants were denied 12.8% (95% CI 12.2% to 13.5%) of the time against 6.5% (95% CI 6.2% to 6.8%) for white applicants in the same 10,189 tracts, a gap of 6.4 (95% CI 5.9 to 6.9) points or 2.0 (95% CI 1.9 to 2.1) times. For Asian applicants the rates were 8.9% (95% CI 8.5% to 9.4%) and 5.7% (95% CI 5.4% to 6.0%) in 5,490 tracts, a gap of 3.2 (95% CI 2.8 to 3.6) points or 1.6 (95% CI 1.5 to 1.7) times. The Black gap is larger than the Hispanic gap, which is larger than the Asian gap. Over each group's own tracts the Black gap is 10.5 (95% CI 9.9 to 11.2) points, so holding the tract fixed changes little. The tracts that hold both cells are a small and selected set, and 27% of their Black cells rest on fewer than 10 decided applications. Requiring at least 10 decided applications from both groups leaves the Black gap at 10.0 (95% CI 9.5 to 10.6) points, and requiring at least 30 leaves it at 10.8 (95% CI 10.1 to 11.6) points.

Figure 4

In the tracts that hold both groups, Black applicants were denied 16.2% of the time and white applicants 6.2%

Denied share of decided home-purchase applications, 2024, for each applicant group and for non-Hispanic white applicants in the same tracts, tracts where both cells are published

Non-Hispanic white applicants in the same tracts
Applicants of the group in the row

Show data

Data for In the tracts that hold both groups, Black applicants were denied 16.2% of the time and white applicants 6.2%
group	hi	lo	rate	tracts	white	white_hi	white_lo
Non-Hispanic Black applicants	16.77	15.67	16.20	5657	6.23	6.50	5.96
Hispanic (any race) applicants	13.53	12.17	12.83	10189	6.47	6.77	6.19
Non-Hispanic Asian applicants	9.37	8.50	8.91	5490	5.70	5.99	5.42

The group's dot lies to the right of the white dot in every row, and the gap is 10.0 percentage points for Black applicants, 6.4 for Hispanic applicants and 3.2 for Asian applicants. Each row pools the applications of the group and of white applicants over the tracts that hold both cells, so the white dot differs across rows. The 95% interval comes from a bootstrap over counties.Note: The same tract holds the neighborhood fixed, not the applicant. The tract file has no credit score, debt-to-income ratio, loan-to-value ratio, loan type or lender for each applicant, so a gap here is not evidence of discrimination by itself and the data cannot say how much of it these items account for.Note: The tracts that hold both cells are 7.4% of all tracts for Black applicants. The tracts differ by row (5,657 for Black applicants, 10,189 for Hispanic applicants and 5,490 for Asian applicants), because a cell exists only where a group has 10 or more applications. Over each group's own tracts the Black gap is 10.5 points.Source: HMDA Loan/Application Register, 2024, tract totals of decided applications; Area Deprivation Index, 2023; full counts, pooled rates

The gap shows no trend with the deprivation of the tract or with the typical applicant income in it. Across tenths of the deprivation index, the same-tract gap between Black and white applicants runs from 8.6 to 10.4 points, its interval is above zero in every tenth, and its slope is 0.00 (95% CI −0.22 to 0.21) points per tenth. Across fifths of tracts by the median income reported on their applications, from $74 thousand or less to $125 thousand or more, it runs from 9.1 to 10.3 with a slope of −0.06 (95% CI −0.33 to 0.20) per fifth (Figure 5). Single tenths differ from one another, so the claim is about the trend and not about every tenth. The ratio of the Black rate to the white rate is lower in the most deprived tenth, 2.0 (95% CI 1.8 to 2.3), than in the fifth tenth, 2.8 (95% CI 2.6 to 3.0). The white rate is higher in the most deprived tenth, so the same gap in points is a smaller multiple.

Figure 5

The same-tract gap between Black and white applicants stays between 8.6 and 10.4 percentage points across tenths of tract deprivation

Black minus white denied share of decided home-purchase applications, 2024, in tracts where both cells are published, by tenth of tracts on the deprivation index and by fifth of tracts on median applicant income

The Black minus white gap is positive in every tenth of tract deprivation

Percentage points of denial rate, Black minus white applicants in the same tracts, by tenth of tracts on the Area Deprivation Index (1 = least deprived)

Gap in denial rate, Black minus white (percentage points)

The gap shows no trend from the least to the most deprived tenth, with the widest interval in the least deprived tenth, where the fewest tracts hold both cells.

Show data

Data for The same-tract gap between Black and white applicants stays between 8.6 and 10.4 percentage points across tenths of tract deprivation, By tract deprivation
conf_high	conf_low	gap	tenth	tracts
13.59	7.64	10.31	1	77
11.12	7.84	9.57	2	335
9.56	7.69	8.61	3	514
10.79	8.83	9.81	4	725
11.14	9.46	10.33	5	809
11.61	9.34	10.45	6	849
10.36	8.55	9.44	7	776
10.85	8.70	9.72	8	628
11.39	9.06	10.19	9	547
10.72	7.92	9.32	10	395

The gap is positive with an interval that excludes zero in every tenth and every fifth, and it shows no trend: the slope is 0.00 points per tenth in the first tab and −0.06 per fifth in the second. The first tab groups tracts by the deprivation index; the second groups the same kind of tracts by the median income that applicants reported, from $74 thousand or less in the lowest fifth to $125 thousand or more in the highest. Each point pools the applications of the two groups over the tracts in its group, with 95% intervals from a bootstrap over counties.Note: The tenths hold near-equal numbers of the tracts with a rank (tracts tied on the rank stay together); the fifths hold near-equal numbers of the 5,657 tracts that hold both cells and a median applicant income, so a fifth groups tracts by the typical income of their applicants and not applicants by their own income. The median income is the median of the incomes reported on the tract's applications.Note: The ratio of the Black to the white rate is 2.8 in the fifth tenth and 2.0 in the most deprived tenth.Source: HMDA Loan/Application Register, 2024, tract totals of decided applications; Area Deprivation Index, 2023; full counts, pooled rates

A tract holds the neighborhood fixed and leaves the applicant free to differ. Applicants in one tract can differ in credit history, debt relative to income, the down payment, the loan type and the lender they chose, and none of these is in the tract file. A study of Boston lenders, published in 1996, added fields for credit history, debt burdens and loan-to-value ratios to the application records and found that the difference in denial rates between minority and white applicants narrowed and did not disappear.[1] The tract file cannot repeat that exercise. Its gaps can be larger or smaller than the gaps among applicants with the same finances, and the data do not say which.

Causal evidenceMatched-pair email correspondence experiment

Whether loan officers respond differently to names that signal race

Hanson, Hawley, Martin and Liu test the step that comes before an application.[2] They emailed licensed mortgage loan originators with a first inquiry about a loan. The inquiries varied the sender's name, chosen to signal a Black or a white sender, and the credit score the sender reported. Because the researchers set the name and the reported score in each inquiry, a difference in how originators replied can be attributed to them. Originators answered inquiries with Black-signaling names less often, and they gave those senders less loan information and sent fewer follow-up messages. The authors compare the size of the name effect on response with the effect of a credit score 71 points lower.

The experiment measures replies to inquiries and ends before any application, so it observes no denial. Its estimates are for the originators and inquiries in the sample.

A person whose inquiry goes unanswered, or is answered with less information, may never apply, and does not appear in a denial rate. That is one way the applications in this chapter differ from the people who looked for a loan.

Grades drawn in the 1930s and denial rates in 2024

Between 1935 and 1940 the Home Owners' Loan Corporation graded neighborhoods in the cities it surveyed from A, labeled best, to D, labeled hazardous.[3] Chapter 4.1, Home value, income and redlining describes the maps and sets the grades beside life expectancy and home values. Here each grade is matched to 2020 tracts. A tract takes the grade that covers most of its area when graded polygons cover at least 50% of it. That gives 10,967 graded tracts in the country, of which 8,748 have a published home-purchase cell and a deprivation rank, in 221 city and state pairs. They are the cities that HOLC surveyed and not a sample of the country. Graded tracts hold 7.3% of the decided applications in this chapter, and their pooled denial rate is 10.3% (95% CI 9.5% to 11.1%) against 8.9% (95% CI 8.6% to 9.2%) in all other tracts with a deprivation rank.

Denial rates rise from 7.9% (95% CI 7.0% to 9.0%) in grade A tracts to 9.4% in B and 10.9% in C, and differ little between C and D, which has 11.3% (95% CI 10.2% to 12.5%). The gap between D and A is 3.4 (95% CI 2.1 to 4.7) points (Figure 6), and the step from C to D is 0.4 (95% CI −0.6 to 1.5). Comparing tracts within the same city leaves the gap at 3.5 (95% CI 2.1 to 4.8), and requiring graded polygons to cover at least 90% of a tract leaves it at 2.8 (95% CI 1.0 to 4.6). Comparing tracts of the same city and the same tenth of the deprivation index, which gives each of 1,008 city and tenth cells its own baseline, removes it. Grade D tracts differ from grade A tracts by 0.0 (95% CI −1.1 to 1.1) points in that comparison, and grade B and C tracts by −0.2 (95% CI −0.9 to 0.5) and −0.4 (95% CI −1.3 to 0.6), all with intervals around zero. In 2024 the differences across grades line up with the deprivation of the tracts today. The comparison does not say whether the maps had a part in how those tracts came to be deprived, and the boundary study in Chapter 4.1, Home value, income and redlining addresses that question for housing outcomes.

Whether the grade gradient differs by applicant group cannot be told from these data. Grade D minus grade A is 1.4 (95% CI 0.2 to 2.8) points for white applicants, 0.1 (95% CI −5.5 to 6.0) for Black applicants and −1.3 (95% CI −7.7 to 4.4) for Hispanic applicants, each over the tracts where the group's cell is published. The Black and Hispanic intervals hold zero and also the all-applicant gap, and the white gap minus the Black gap is 1.3 (95% CI −5.0 to 7.4) points. Grade A tracts have few cells for these groups. The mix of applicants also changes with the grade. Published Black and Hispanic cells account for 7.8% (95% CI 5.0% to 11.0%) of decided applications in grade A tracts and 15.0% (95% CI 11.2% to 18.9%) in grade D tracts, a lower bound because withheld cells are not counted, and both groups have higher denial rates, so part of the all-applicant gradient reflects who applies in each kind of tract.

Figure 6

Denial rates rise from 7.9% in grade A tracts to 11.3% in grade D tracts, and no grade differs detectably from grade A among tracts of the same city and the same tenth of the deprivation index

Denied share of decided home-purchase applications, 2024, in tracts graded on the 1930s HOLC maps: by grade, as a difference from grade A under three adjustments, and by applicant group

The denial rate rises from grade A to grade C and differs little between grades C and D

The dots step to the right from grade A to grade B and from grade B to grade C, grade D sits close to grade C, and the intervals for grades A and D do not overlap.

Show data

Data for Denial rates rise from 7.9% in grade A tracts to 11.3% in grade D tracts, and no grade differs detectably from grade A among tracts of the same city and the same tenth of the deprivation index, By grade
conf_high	conf_low	estimate	grade	label	tracts
8.96	6.98	7.91	A	Grade A	650
10.27	8.47	9.37	B	Grade B	2003
11.81	10.01	10.91	C	Grade C	4008
12.50	10.21	11.32	D	Grade D	2087

The first tab shows the rate rising from grade A to grade C, with grade D close to grade C; the second shows the differences from grade A shrinking to nothing detectable when tracts are compared within the same city and the same tenth of the deprivation index; the third shows each applicant group's rates by grade, where grade A rests on few tracts for the Black, Hispanic and Asian groups and their intervals are wide. A group appears only where a grade rests on at least 30 tracts. The 95% interval comes from a bootstrap over counties in the first and third tabs and from county-clustered regression errors in the second.Note: A tract takes a grade when graded polygons cover at least 50% of its area, and the grade that covers the most of it. The tracts are the 8,748 graded tracts with a published cell and a deprivation rank, in 221 city and state pairs. With a stricter rule, graded polygons covering at least 90% of a tract, the gap between grades D and A is 2.8 points. The comparison is descriptive: the grades followed the housing, incomes and populations of the 1930s, and the HOLC cities are not a national sample.Note: The adjusted differences come from weighted regressions of the tract denial rate on grade, weighted by decided applications, with errors clustered by county, on the 8,748 graded tracts. The third adjustment gives each of 1,008 city and tenth cells its own baseline, so a grade is compared with another grade only inside a cell that holds tracts of both.Source: HMDA Loan/Application Register, 2024, tract totals of decided applications; Mapping Inequality HOLC maps (1935-1940) overlaid on 2020 tracts; Area Deprivation Index, 2023; full counts, pooled rates

The Cook County map shows the same data at tract scale (Figure 7). Of its 1,331 tracts, 914 decided at least 20 home-purchase applications and are shaded, and 417 are gray, each with its reason in the tooltip. The Chicago map graded 759 of the county's tracts, all of them in the city, and the ungraded tracts include parts of the county that the map did not reach. Among the shaded tracts, 53.0% of those in the top fifth of denial rates and 57.7% of those in the bottom fifth were graded, a difference of −4.7 (95% CI −14.8 to 5.6) points. That compares graded with ungraded tracts, which is not the comparison of grade D with grade A, and the interval is too wide for these data to say whether a tract's place in the graded part of the county goes with a higher or a lower denial rate. Within the graded part, the pooled rate is 5.4% (95% CI 2.5% to 8.9%) in grade A tracts, which rest on 11 tracts, 12.3% (95% CI 10.1% to 14.8%) in B, 11.0% (95% CI 10.0% to 12.0%) in C and 9.8% (95% CI 8.4% to 11.3%) in D, against 11.4% (95% CI 10.8% to 12.1%) in tracts with no grade.

Figure 7

In Cook County, 53.0% of the tracts in the top fifth of denial rates and 57.7% of those in the bottom fifth lie in the part of the county that the 1930s map graded, a difference these data cannot tell from zero

Denied share of decided home-purchase applications, 2024, in Cook County tracts with at least 20 decided applications, and the HOLC grade covering most of each tract

Loading tract outlines

Denied share of decided home-purchase applications, 2024 (%)

No data (0 tracts)

Select a class to show only those places.

Find a tract

Hover or tap a tract for its value, or search by name. Zoom with the buttons on the map, then drag to move it.

The first view shades each tract by its denial rate; gray tracts have no rate, and the tooltip says why. The second view colors each tract by its 1930s grade, with gray for tracts that graded polygons cover less than halfway, which include parts of the county that the Chicago map did not reach. Comparing graded with ungraded tracts is not comparing grade D with grade A, and grade A rests on 11 tracts.Note: 914 of the 1,331 tracts have a rate and 417 do not, because the tract has fewer than 20 decided applications, fewer than 10 applications in all (withheld), or no row in the file. 759 tracts have a grade, all of them in Chicago. The rate in the tooltip carries a 95% interval from the Wilson formula, which treats applications as independent, so it is narrower than a bootstrap over tracts would give.Note: The share graded in the top fifth minus the share in the bottom fifth is −4.7 percentage points, with an interval from a bootstrap over tracts that includes zero. Pooled over the shaded tracts, the denial rate is 5.4% in grade A tracts, 12.3% in B, 11.0% in C, 9.8% in D and 11.4% in tracts with no grade, with intervals from a bootstrap over tracts.Source: HMDA Loan/Application Register, 2024, tract totals of decided applications; Mapping Inequality HOLC maps (1935-1940) overlaid on 2020 tracts; full counts, pooled rates

What these data cannot show

Applications are not the population. A denial rate is a share of the applications that reached a decision, and who applies differs across tracts and groups. People who expect a denial, who are steered to another loan or who cannot get an answer to an inquiry do not appear, and withdrawn applications leave the rate. The file covers one year and the lenders above the reporting thresholds.

The tract totals hold no credit score, debt-to-income ratio, loan-to-value ratio, loan type or lender for any applicant. A gap between groups, even inside one tract, is therefore not evidence of discrimination by itself and not a measure of how much treatment differs. The application-level HMDA records carry some of those fields and up to four denial reasons for each denied application. Lenders report those reasons themselves, they are not audited, and the tract aggregate does not keep them.

The groups are reported groups. The ethnicity and race in HMDA come from the applicant where the applicant answers and from fallback rules where not, and the four groups leave out other and unreported applicants. A group's cell exists only in tracts with 10 or more of its applications, so group comparisons cover the tracts with more applicants of that group and not a random sample of tracts.

The intervals resample counties. Applications at one lender share its rules and cross county lines, so the intervals do not capture correlation within a lender and may be narrower than the uncertainty. The tenths and grades describe tracts and the applications reported in them. They are not the experience of one borrower, and a pattern across places need not hold for the people inside them (ecological inference). The deprivation index understates deprivation in high-cost cities and is built partly from income and housing costs, which are also what a lender weighs.

The grades followed the housing, incomes and populations of the 1930s, so a tract's grade stands for a mix of what the neighborhood was and what happened after. The adjusted contrast compares tracts of the same city and the same tenth of the index and cannot say whether the maps shaped today's deprivation. A city that HOLC did not survey is absent from the comparison, which does not mean it had no redlining. Tooltip intervals on the Cook County map treat applications as independent and are narrower than a bootstrap over tracts would give.

Data sources

HMDA mortgage lendingHome Mortgage Disclosure Act Loan/Application Register, 2024, first-lien site-built one-to-four-unit principal-residence applications aggregated to 2020 tracts by loan purpose and applicant group; cells under 10 applications withheldYears: 2024 · Weights: none (counts of decided applications, pooled over tracts) · License: Public domain (U.S. federal)
Area Deprivation IndexArea Deprivation Index 2023, national rank by tractYears: 2023 release (version 4.0.1; built on ACS 2019-2023) · Weights: none (tenths hold near-equal numbers of tracts) · License: Free for non-commercial use; cite the Neighborhood Atlas and Kind and Buckingham
Mapping Inequality (HOLC maps)Mapping Inequality HOLC residential security maps, 1935 to 1940, graded polygons overlaid on 2020 tractsYears: 1930s-1940s maps (2023 digitization) · Weights: decided applications · License: CC BY-NC-SA 4.0 (cite Nelson and coauthors)
Published estimates from cited studiesOne comparison quoted from the abstract of Hanson, Hawley, Martin and Liu (2016)Years: as reported in each study · Weights: as reported in the paper · License: Quoted from published research for commentary

All data sources

Methods

This chapter compares denial rates on home-purchase applications decided in 2024 across census tracts, for all applicants and for four applicant groups. The file covers lenders above the HMDA reporting thresholds and has no loan type; withdrawn and incomplete applications are not in the rate, and a tract, purpose and group cell with fewer than 10 applications is missing, so a comparison between groups covers only tracts where both cells are published. The HOLC grades come from maps drawn in the 1930s and are compared with 2024 lending, and the Area Deprivation Index that ranks tracts is built partly from income and housing costs.

Which applications are included

The staged HMDA file holds 2024 first-lien applications on site-built homes with up to four units that applicants intend as a principal residence, from lenders above the HMDA reporting thresholds, aggregated to 2020 census tracts by loan purpose (home purchase, refinance, cash-out refinance) and by applicant group (all, and the applicant's reported non-Hispanic white, non-Hispanic Black, Hispanic of any race and non-Hispanic Asian).

The chapter uses home-purchase applications in the 50 states and DC. Puerto Rico is left out. Loan type is not in the file.

Figure explained here: Figure 1

How the denial rate is defined

The denial rate is denied applications divided by applications that were originated or denied (decided), as a percent, summed over tracts: a pooled rate, never an average of tract rates.

Applications that were withdrawn or left incomplete, loans approved but not accepted and purchased loans are not in the rate.

Cells with few applications are missing

When the file was staged, a tract, purpose and group cell with fewer than 10 applications (of every action taken) was set to missing; a missing cell leaves both the numerator and the denominator and is never imputed or back-calculated.

How tracts are grouped

Tenths hold near-equal numbers of tracts ranked by the Area Deprivation Index national rank (tracts tied on the rank stay together). The fifths of tract applicant income hold near-equal numbers of tracts that have published cells for Black and white applicants.

Figures explained here: Figure 2 and Figure 5

HOLC grades and the Cook County map

Graded polygons (A to D) are intersected with 2020 tracts in an equal-area projection; a tract with at least 50 percent of its area graded takes the grade covering most of it and the city of the polygon covering most of it (the rule of chapter 4.1); the grade D minus grade A gap is also shown for a stricter rule of 90 percent coverage.

Grade comparisons are weighted regressions of the tract denial rate on grade, weighted by decided applications, with errors clustered by county, with no baseline, with a separate baseline for each city, and with a separate baseline for each city and tenth cell.

On the Cook County map, tracts with at least 20 decided applications are shaded; the tooltip interval is a Wilson interval on the counts.

Figures explained here: Figure 6 and Figure 7

Intervals, trends and the tract sample

Intervals are percentile intervals from a bootstrap that resamples whole counties with replacement (2,000 replicates, seed 12345, the same draws for every estimate so a difference gets one interval).

The trend of the same-tract gap is the least-squares slope of the gaps of the tenths (or fifths) on their number, recomputed in every bootstrap draw.

The scatter shows a random sample of 2,000 tracts with at least 30 decided applications, drawn with seed 12345.

Figures explained here: Figure 4 and Figure 2

References

Munnell, Alicia H.; Tootell, Geoffrey M. B.; Browne, Lynn E.; McEneaney, James (1996). Mortgage Lending in Boston: Interpreting HMDA Data. American Economic Review 86(1): 25-53.
Hanson, Andrew; Hawley, Zackary; Martin, Hal; Liu, Bo (2016). Discrimination in Mortgage Lending: Evidence from a Correspondence Experiment. Journal of Urban Economics 92: 48-65. doi:10.1016/j.jue.2015.12.004
Nelson, Robert K.; Winling, LaDale; Marciano, Richard; Connolly, Nathan; et al. (2023). Mapping Inequality: Redlining in New Deal America. American Panorama, ed. Robert K. Nelson and Edward L. Ayers, Digital Scholarship Lab, University of Richmond. dsl.richmond.edu/panorama/redlining/

How to cite

@misc{singh2026mortgagedenials,
  author = {Singh, Vishal},
  title = {Mortgage denials},
  year = {2026},
  url = {https://vishalsingh.org/zipcode/mortgage-denials},
  note = {In ZIP Code Destiny. Published 9 October 2026.}
}